AI Infrastructure
AWS and NVIDIA Plan 2 Million More GPUs for 2027-28
AWS and NVIDIA announced on August 26 that AWS plans to deploy two million additional Blackwell Ultra, Rubin, and Rubin Ultra GPUs across its global infrastructure in 2027-28. The number is a future capacity plan, not current installed, orderable, utilized, or independently audited supply.
Citation-ready: AWS plans to deploy two million additional NVIDIA Blackwell Ultra, Rubin, and Rubin Ultra GPUs across its global infrastructure during 2027 and 2028.
What happened and why it matters
No. The announcement describes a 2027-28 plan across three GPU generations and a broader co-engineering agreement; availability needs dated regional instance, quota, delivery, power, networking, and utilization evidence.
Official joint AWS and NVIDIA press release
Primary reference: AWS and NVIDIA infrastructure expansion announcement. Kaleido Field checked the event date and the article's attributed facts against this source.
| Source date | August 26, 2026 |
|---|---|
| Checked by Kaleido Field | August 31, 2026, 09:04 CST |
| Source function | current cloud-capacity analysis separating announced units, architecture mix, deployment window, installed and orderable capacity, utilization, power, region, and customer economics |
The architecture mix matters
The plan spans Blackwell Ultra, Rubin, and Rubin Ultra rather than one interchangeable unit. A count alone does not describe memory, network, rack density, power, software compatibility, or the work each system can complete.
A useful capacity ledger needs model, configuration, rack, region, order date, installed date, service date, quota, useful throughput, outage history, and retired capacity.
Capacity can be announced before its dependencies exist
Large GPU deployments depend on manufacturing, packaging, memory, networking, buildings, grid connections, cooling, permits, and service operations. The joint release expands work across several of those layers but does not publish a construction or delivery schedule.
Kaleido Field will separate planned units, delivered units, commissioned instances, customer-accessible quota, and utilized compute whenever later evidence appears.
Evidence boundary
Official plan: unit count, architecture families, deployment window, global scope, and named areas of collaboration. Company demand and performance claims: customer need, platform scale, EMR comparisons, and strategic value. Not established: purchase completion, delivery schedule by quarter or region, installed racks, power and network readiness, customer quotas, utilization, price, carbon or water impact, or independent audit.
FAQ
When are the GPUs planned?
AWS says 2027-28.
Which GPU families are included?
Blackwell Ultra, Rubin, and Rubin Ultra.
Are they available now?
The release announces future deployment; it does not report two million installed or orderable GPUs today.